Chris
Snijders
© 2012 Chris Snijders Contact Me
The point

Scientific studies repeatedly show that decisions made by computers are often just as good or better than the decisions made by human experts. Early studies primarily focused on clinical applications (showing for instance the computer's ability to diagnose certain diseases), but later studies found similar results when it comes to financial and business matters, including managerial decisions. In the scientific literature the comparison of man-made vs. machine-made decisions is called "clinical vs statistical prediction".
In my own research I have repeatedly tested, often together with my colleague
Frits Tazelaar, how managerial decisions made by computer models compare to the decisions made by managers with various degrees of expertise. To cut a longer story short: the computer model easily wins.


Can computers improve decisions in my organization?

More often than you might think. There are a couple of rules thumb on when computers are likely to do at least as good as humans in a business context:
+
It must be a decision that can be quantified. That's obvious. You cannot tell a computer to design your marketing strategy and present it to the board, at least not yet. Instead you want a model to look at questions such as: how many people should we put on this sales team? How much should we invest in trying to make our contract water-tight? Should I go to my supervisor to have him check this? Which supplier should we choose?
+
It must be a decision about which you have at least some quantified data available (you should at least be able to collect such data in a reasonable amount of time). Also pretty obvious: if there are no data available to predict from, no computer model can be made.
+
It must be a decision that occurs relatively frequently. The more data you have (or can collect in a reasonable amount of time), the better the model will be able to predict. Moreover, the more often this kind of decision occurs, the better the computer can show its superior performance. Note: in my own research a model with only a handful of predictor variables easily outperforms experts who have more than twice as much data and a lifetime of experience available. So yes, there are requirements on having data, but no massive and complicated data mining is necessary.

A business challenge
In many business decisions, a computer model will outperform a human. In other words: many managers are now making decisions that they could just as well leave to a computer model. In fact, I am prepared to put my money where my mouth is.
If you are willing to objectively test whether computer models can do your business or organization any good, I am willing to invest my time and effort in both designing the test and preparing a proper computer model that can make the decision. The only thing I want in return is to be able to use the data of the comparison for scientific purposes. You can't loose. Interested? Send me an email.